So I am implementing a feature where I want to chat with the database, Database is a RDS Postgresql. I have a mid size database so there are too many tables and columns so I cannot afford to send the database schema to the LLM for every prompt to generate a response query, how can I optimize here?
One thing I thought was as per RBAC we can create roles that can access specific tables and columns in the database this will reduce the schema input size. But still we are using tokens repetitively to provide same information again and again.